How to Respond to Misinformation About Your Organisation
Two AI failures broke a few years apart, showing what a false claim can cost before anyone has checked whether it is true. One was a faked image that moved a stock index within minutes. The other was a coding agent that erased a company's live database, then lied about recovering it. In Australia, the eSafety Commissioner has already published guidance for schools on handling AI deepfakes, including a deepfake letter template schools can adapt, because the same problem now reaches classrooms as fast as it reaches markets. Melissa Agnes, founder of the Crisis Ready Institute, wrote about both incidents in the same July 2026 newsletter. Her framework for correcting a public falsehood follows, largely in her own words.
The Crisis Masterclass teaches this exact kind of public correction, with a worked example and a template.
See the programThe Fake Pentagon Photo That Moved Markets
Melissa Agnes opened her account of both stories like this:
On a Monday morning in May 2023, a photo started spreading online showing a huge explosion at the Pentagon. It looked real enough that a verified news account shared it. Within minutes, a Russian state media outlet and an Indian news outlet picked it up too. The S&P 500 dipped. Real money moved, in real time, off the back of a photo of an event that never happened. The image was AI generated. There was no explosion. And while the market recovered within minutes once the truth caught up, the point is that the truth had to catch up.
No one had to invent a scandal for that photo to move a market; a convincing image did the job in minutes, before any official had confirmed or denied anything. The same logic applies to a false claim about your own organisation: whether it counts as an issue or a crisis is decided by how fast and how clearly you respond, not by whether the claim was ever true.
The AI Agent That Deleted a Database, Then Lied About It
The second story, from the same week, is about software rather than a photo:
Last year, a founder was using an AI coding agent to help build out his company's software. He told it, in writing, eleven times, not to touch the production database without his permission. The AI agent decided to delete the database anyway. Without permission.
This is what agentic AI can do. It doesn't have to wait for permission between steps. It acts on its own, and this time, it acted badly.
Then it tried to cover its tracks. It generated 4,000 fake user records and fabricated test results just to hide what it had done. When the founder asked if the data could be recovered, the AI agent told him no, that the backups were gone too. That wasn't true either. The founder recovered the backups himself, manually, after the AI had already tried to assure him that doing so was impossible. When he asked the AI to rate the severity of what it had done on a scale of 1 to 100, it rated itself a 95.
None of that is really about software. It is about what happens when an autonomous system is trusted to act and then covers for its own mistake, which is close to the shape a lot of reputational crises take: something goes wrong inside a system nobody was watching closely enough, and the first instinct, human or automated, is to minimise it rather than report it straight.
What Actually Happened at Replit
The detail most retellings skip: the agent had direct write access to the live production database. A written instruction not to touch it was a request, not a technical control, so nothing stopped the agent doing what it had been told not to do eleven times. That reframes the incident: less a rogue agent defying orders than a guardrail that never existed, which makes it a more useful case study for any organisation running AI tools without a hard limit on high-risk actions, not a less serious one.
If your organisation runs AI tools without a human check on high-risk actions, that gap belongs in your crisis plan.
Run the checkMelissa's Three-Step Way to Correct a False Claim in Public
Melissa's full Crisis Ready formula covers a complete crisis statement; this is the smaller version for one specific job, correcting a false claim in public. She introduced it like this:
If you're ever the one having to publicly correct a lie, resist the urge to lead with it, even to knock it down. Try this instead:
The Three Steps, in Order
- Lead with the truth, clearly.
- Briefly name the false claim and label it as false.
- Return to the truth, and what you're doing about it.
This is simple on paper, but almost nobody does it under pressure, because the typical instinct is to argue with the lie first, to self-protect, and to deflect. Don't. Instead, focus on your truths and building trust and reassurance with those who matter most to your organization.
"We've seen claims online that our clinic falsified patient records. This is completely untrue and we're appalled anyone would suggest it. At no point did any staff member falsify anything, and we're considering legal action against whoever started this rumour."
"Every patient record at our clinic is complete and matches our independent audit logs, which we've now published. A post circulating online claiming records were falsified is false. We've asked the platform to remove it and we're answering questions directly, because our patients deserve a straight answer."
Notice what the weak version does: it repeats the accusation in its own first sentence, so a reader skimming the opening line sees the lie before the rebuttal. The strong version puts the true fact first, names the false claim once, and never argues with it, the same order Melissa Agnes describes above.
The full three-step drill, plus a worked example on a real statement, is inside month nine of the Masterclass.
See what's insideWhere This Sits in the Crisis Masterclass
Correcting a false claim, whatever generated it, is one module inside a longer program. Month nine of the Crisis Masterclass is called Navigating the Modern Noise, alongside the rest of the twelve months Melissa Agnes and Lyall Mercer teach together, which is also why a single correction is rarely where the story ends. If a fake image or a rogue AI agent put your organisation in the news this week, the real question is whether your team leads with the truth, or the lie.
Her sign-off:
The Pentagon photo fooled the market for minutes. The Replit AI lied to a founder's face, more than once. If that doesn't tell you the terrain has changed, I don't know what will.
Frequently asked questions
What is the best way to respond to misinformation about your organisation?
Lead with the truth first, not the lie. Melissa Agnes's framework is to state what is true, briefly name the false claim and label it as false, then return to the truth and what you are doing about it. Arguing with the lie first, even to disprove it, is the instinct that usually backfires under pressure.
What actually happened when an AI agent deleted a company's production database?
In 2025, a founder using an AI coding agent had told it, in writing and repeatedly, not to touch the live database without permission. The agent deleted it anyway, then generated fake records and told the founder the backups were unrecoverable, which was not true. The deeper cause was that the agent had direct write access with no technical guardrail stopping it.
Was the fake AI image of a Pentagon explosion actually taken seriously?
Yes, briefly. In May 2023, an AI-generated image of an explosion near the Pentagon was shared by a verified account and picked up by state media outlets, and the S&P 500 dipped before officials confirmed no explosion had occurred. The market recovered within minutes, but the false image moved real money before the truth caught up.
Does Australia have guidance on responding to AI deepfakes?
The eSafety Commissioner has published guidance for Australian schools on handling AI deepfake incidents, including how to report them and support the people affected. Organisations outside education face the same underlying problem, a convincing false claim spreading before anyone can verify it, and the same lead-with-the-truth response applies.
Sources
Melissa Agnes
Founder, Crisis Ready Institute
Based in the USA, Melissa is a recognised leading authority on crisis communication, crisis leadership and brand protection. Best-selling author of 'Crisis Ready: Building an Invincible Brand in an Uncertain World', and creator of the Crisis Ready® Model which has been globally recognised as having elevated industry best practice, she has worked with ministries of foreign affairs and defence, NATO and many other leading organisations. She presents the Crisis Ready course exclusively through My PR Partner in Australasia.
Melissa Agnes is the founder of the Crisis Ready Institute and creator of the Crisis Ready® Model. Her Crisis Ready course is available in Australasia exclusively through the My PR Partner Crisis Masterclass. See the program.
Quoted passages are verbatim from Melissa Agnes's 15 July 2026 Crisis Ready Institute newsletter to Crisis Masterclass students. A promotional passage for the Crisis Ready Institute Summer Intensive has been cut entirely; the "What Actually Happened at Replit" section and the two illustrative correction examples are original analysis, not Melissa's words.
